Here is the abstract you requested from the RF_2014 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|Passive Macromodeling Methodology for TSV and Bondwire Interconnects|
|Keywords: macromodeling, signal integrity analysis, noise analysis|
|A crucial element in any Integrated Circuit (IC) system design is the interconnects which are responsible for the power delivery and signal transmission, within the IC itself, and those channels which connect to the IC packaging. The demand of digital systems to provide gigabit data rates has brought about engineering challenges related to reliably convey high speed signals within the chip, and sending these signals beyond the IC packaging. An example of an emerging IC interconnect element is the Thru-Silicon-Via (TSV), while the bondwire is a commonly encountered interconnect between the IC and the packaging. Macromodeling is a methodology employed with the goal to perform time-domain SPICE analysis of these interconnects, using their frequency transfer characteristics to extract a SPICE equivalent circuit, to predict and mitigate their noise performance behavior to improve signal transmission. Generally, electromagnetic (EM) simulators are commonly used for electromagnetic compatibility (EMC) and signal integrity (SI) analysis interconnects. The accuracy of such analysis depends on the macromodels used for emulating the frequency transfer characteristic of the interconnect. These models should be broadband and preserve the physical properties of the materials, such as causality and passivity. The passivity constraint associated with macromodeling is one of the more challenging requirements to satisfy, which is a guarantee of the positive realness of the interconnect model across all frequencies, or that there is no energy gain performed by the model. In this paper, this is performed by the use of non-negative least squares fitting, which guarantees positive real values at all points in the model corresponding to data values. In this effort, single-input single-output macromodel analyses of the bondwire and TSV are demonstrated, and show good agreement between model and data, while achieving model-order reduction, and satisfying the passivity and causality requirements for the respective macromodel.|
San Diego State University
San Diego, CA